For cloud computing, particularly products that provide virtual machine services, as the scale of physical machines continuously increases, different batches of physical machines will be purchased, and there will be multiple physical machine models in a same cluster. During virtualization implementation, the virtualization of the memory and magnetic disks is not greatly affected by physical machines; however, in the case of different types of central processing units (CPUs), when a virtual machine needs to be live-migrated across physical machines (that is, saving/restoring the virtual machine, generally referring to the running state of the entire virtual machine being completely saved, and being quickly restored on the original hardware platform or on a different hardware platform; after restoration, the virtual machine still run smoothly, without the user perceiving any difference), there is a problem of instruction set incompatibility, which is also an important problem to be solved for live-migrating virtual machines in a heterogeneous cluster.
There are mainly two solutions to the problems of CPU model consistency, performance, and live migration compatibility of a large-scale cluster:    1) Ensure that the cluster uses same processors, and a guest machine (that is, virtual machine) adopts a host model (host_model) or host passthrough model as its CPU model (CPU_model).    2) Use a custom model (custom) that is supported by all physical machines in the cluster.
The first solution is to ensure the performance and compatibility features by means of hardware consistency; while the second solution takes into consideration the requirements of highly heterogeneous clusters, and causes a greater performance loss.